Countries that treat electricity, land, water and digital infra as separate policy domains will struggle to host advanced computing at scale. Integration to shape where intelligence is built, deployed, governed.
Countries that treat electricity, land, water and digital infra (Photo: Pinterest)
The debate on artificial intelligence and energy is still framed too narrowly around how much power AI consumes. That question matters less than why AI creates such acute stress for electricity systems, where that stress materialises, and how it propagates through grids, prices and industrial policy. Once those mechanisms are unpacked, the issue looks less like a demand shock and more like a systems problem, particularly for countries such as India.
Recent analytical work points in the same direction even as headline estimates vary. The IMF Working Paper Artificial Intelligence and Electricity: A Macroeconomic Framework shows that AI-related electricity use behaves differently from conventional industrial demand. Its macroeconomic impact is driven not by aggregate consumption but by binding constraints in transmission, reliability and peak availability. In scenarios where generation expands but networks and firm capacity do not, electricity prices rise sharply, emissions increase, and the productivity gains from AI are diluted.
Three Structural Features Explain This Dynamic
The first is spatial concentration. Advanced AI workloads are organised around tightly coupled clusters of accelerators, where performance depends on low latency, high bandwidth and physical proximity. This produces nodal electricity demand at the scale of hundreds of megawatts and, in some cases, gigawatts. The IMF modelling shows that price effects are dominated by local congestion rather than national scarcity. This is why electricity systems with ample installed capacity can still experience stress. In India, close to two-thirds of operational data centre capacity is concentrated in Mumbai and Chennai, both regions with limited grid headroom, high land prices and rising water stress. Power may exist elsewhere, but it is not where compute needs to be.
The second feature is temporal rigidity. AI infrastructure is capital-intensive, with utilisation rates that must remain high to justify investment. Unlike many industrial loads, these facilities cannot easily modulate consumption during peak hours without undermining their economics. S&P Global estimates that electricity demand from Indian data centres will grow at roughly 25 to 30 per cent annually through the decade, compared with overall demand growth of around 5 per cent. More importantly, this demand is inflexible and persistent, increasing the system's peak load and reducing the scope for conventional demand response.
The third is the limited offset from efficiency. Power usage effectiveness continues to improve, and chip-level energy efficiency rises steadily. But the scale effect dominates. More capable models, more users and more intensive workloads overwhelm incremental gains. The IMF paper incorporates plausible efficiency improvements and still finds that electricity prices will rise by close to 10 per cent by 2030 in scenarios where transmission expansion and firm generation lag. This explains the wide dispersion in forecasts of AI electricity use but the consistent conclusion that infrastructure constraints, not efficiency, determine outcomes.
For India, These Dynamics Intersect With Existing Structural Weaknesses
Generation capacity in aggregate is not the primary constraint. India is adding renewables at scale, but intermittent supply does not meet the reliability requirements of dense computing clusters without storage or firming. The IMF framework emphasises reliability-adjusted capacity rather than nameplate additions. This distinction is critical. A grid rich in solar and wind but short on storage, gas, hydro flexibility or nuclear baseload struggles to support high-availability loads even if headline capacity targets are met.
Transmission and distribution are more binding. Power is often available in principle but cannot be delivered to suitable sites within commercial timelines. Inter-state transmission, urban substations and last-mile upgrades lag industrial demand. The IMF modelling identifies underinvestment in grids as the single largest amplifier of price volatility and emissions when AI deployment accelerates. India's experience with delayed industrial corridors and stalled generation projects underscores how coordination failures, not resource scarcity, become decisive.
Water adds a further constraint. Despite advances in liquid cooling, large data centres remain water-intensive. Moody's projects India's per capita water availability to fall to around 1,400 cubic metres by the early 2030s, a threshold associated with chronic stress. Concentrating compute infrastructure in already stressed urban regions raises operational, environmental and political risks that electricity planning alone cannot resolve.
Policy Implications
The policy implications are clear and uncomfortable. First, differentiation matters. Not all AI activity imposes the same infrastructure burden. Training frontier models is the most power and capital-intensive segment. Inference, applied AI and sector-specific deployments deliver significant productivity gains with lower peak power requirements. The IMF analysis shows that an economy can capture much of AI's growth dividend even if the most energy intensive training migrates, provided downstream deployment remains domestic.
Second, electricity planning cannot remain siloed. Data centres, grids, generation, land and water must be planned together. Pre-zoned compute corridors with dedicated transmission, firm power contracts and explicit water provisioning reduce uncertainty and system costs. Absent such coordination, private capital will arbitrage toward jurisdictions with faster permitting and clearer timelines, exporting emissions and strategic control in the process.
Third, decarbonisation sequencing matters. The IMF scenarios show that accelerating AI deployment without parallel investment in grids and firm capacity produces worse climate outcomes than a more balanced approach. Delaying coal retirement or expanding gas capacity may be environmentally undesirable, but unmanaged leakage and higher system emissions are worse.
The deeper lesson is institutional. AI exposes the limits of fragmented infrastructure governance. Countries that treat electricity, land, water and digital infrastructure as separate policy domains will struggle to host advanced computing at scale. Those that integrate them will shape where intelligence is built, deployed and governed. For India, the constraint on AI ambition is unlikely to be talent, capital or algorithms. It will be the credibility of the power system as a platform for high-reliability, location-specific demand. In that sense, AI is not primarily an energy challenge. It is a test of state capacity.
Aditya Sinha (x:@adityasinha004) writes on macroeconomic & geopolitical issues.
